JPH03123285A - Vector detector - Google Patents

Abstract

PURPOSE:To improve the accuracy of detection by calculating a second trial vector group so that the trial vectors become dense in the same direction as the dynamic vectors calculated from a first trial vector group, and using it to execute the search of the dynamic vector. CONSTITUTION:A dynamic vector calculation part 1 detects the dynamic vector, and measures the correlation of a frame pixel value and the previous pixel value whose position is shifted by using the trial vector, in units of the divided blocks by using an evaluation function. And, the trial vector with the highest correlation is outputted as the dynamic vector. Next, by using the dynamic vector value, the second trial vector group which is spread in a prescribed area is selected from the first trial vector by a trial vector group selecting means. And, the dynamic vector generated between the former frame and the present frame is calculated with a calculating means 1 at every processing block of the present frame by using the second trial vector group. When this is repeated, and the dynamic vector is searched, the accuracy of detection can be improved.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention The present invention relates to a motion vector detection device used in a motion compensation predictive coding device for television signals and the like.

2. Description of the Related Art In recent years, motion-compensated predictive coding devices have been widely used in fields such as high-efficiency coding of moving images. The principle of motion-compensated predictive coding uses the correlation between frames of a moving image to determine the image movement (called a motion vector) between the current frame and the previous frame, and then calculates the motion of the previous frame based on this motion vector. The current frame image is predicted from the reproduced image, a prediction error is found between this and the current frame image, and the prediction error information and motion vector information are encoded. In the motion compensated predictive interframe coding device, the higher the accuracy of motion vector detection, the lower the prediction error, and therefore the higher the encoding efficiency.

The method for finding the motion vector between frames is to divide the frame into blocks of a predetermined size, calculate the correlation with the image of the previous frame for each block, and find the position with the strongest correlation between that block and the image of the previous frame. A flock matching method is well known in which the positional relationship of blocks is determined as a motion vector.

As a block matching method, JP-A-55-15878
The "interframe coding device" described in Publication No. 4 (hereinafter referred to as conventional example 1) and the "interframe coding device" described in Japanese Patent Application Laid-Open No. 143776/1982 (hereinafter referred to as conventional example 2) are known. It is being Conventional Example 1 is a method in which motion vector detection is performed in N stages (N22). An example of detection in three stages is shown in FIG. In the first stage, Figure 11 (
As shown in ○, ·), the evaluation function (sum of squares of difference signals, sum of absolute values of difference signals, etc.) is calculated for the nine trial vectors.
Calculate the value of and find its minimum value. If (・) is the minimum among the nine trial vectors, in the second stage, the first
As shown in Figure 1 (Δ, M), evaluation function values are calculated for eight trial vectors placed in the vicinity of the trial vector (·) in the same manner as in the first stage, and the minimum value thereof is determined. If the minimum value is (mu), in the third stage, as shown in Figure 11 (mouth, ■), (mu) is the center,
In this method, evaluation function values are calculated for eight trial vectors placed in the vicinity, the minimum value thereof is determined, and the trial vector (■) having the minimum value is set as the motion vector of the block.

Conventional Example 2, like Conventional Example 1, is a method of dividing motion vectors into multiple stages, but in order to detect vectors in a wider range than Conventional Example 1, motion vector detection is performed up to one frame earlier. In this method, a first stage shift vector group is determined using the motion vector of the block as an initial value, and motion vector detection is performed. FIG. 12 shows an example in which the process is carried out in two stages.

If the vector (・) of (4, -4) is the motion vector of the block in the previous frame, as shown in Figure 12,
The eight vectors (Δ, M) and the initial vector (·) in the vicinity of (4, -4) become the first-stage trial vectors. Evaluation function values are calculated for these nine trial vectors, and the minimum value thereof is determined. When the trial vector (mu) is the minimum value in Fig. 12, in the second stage, as shown in Fig. 12, eight trial vectors (mouth, ), the evaluation function is calculated in the same way, its minimum value is found, and the trial vector (■) having the minimum value is set as the motion vector of the block.

Problems to be Solved by the Invention In the multi-stage motion vector detection method shown in Conventional Example 1 and Conventional Example 2 above, the number of trial vectors for detection is smaller than the total number of trial vectors within the detection area. However, since the number of trial vectors in the first stage is small, if a trial vector in a direction different from the actual image movement is selected in the first stage, a correct vector value cannot be obtained.

To prevent this, the number of trial vectors in the first stage must be increased. In this case, the advantage of multiple stages is reduced. Further, when this method is implemented as a device, a complicated control means for performing multi-step search is required, and there is a problem that the device as a whole becomes complicated.

In view of the above problems, the present invention provides a motion vector detection device that can perform detection in one step and has high detection accuracy.

Means for Solving the Problems In order to achieve the above object, the technical solution of the present invention firstly includes block dividing means for dividing one frame of a television signal into blocks of a predetermined size, and trial vector group storage means for storing L first trial vector groups existing in an area of ±2M pixels and ±2N lines in the vertical direction; and motion vector value storage means for storing the motion vector value of the previous frame. Then, using the motion vector value of the previous frame, select P second trial vector groups distributed within an area of ±M pixels in the horizontal direction and ±N lines in the vertical direction from the first trial vector group. The trial vector group selection means selects motion vectors generated between the previous frame and the current frame for each processing block of the current frame by using a second trial vector group in a horizontal direction of ±M around the processing block.
The pixel is provided with calculation means for calculating within a range of ±N lines in the vertical direction. However, M and N are positive integers,
Let L be a positive integer smaller than (4M+1) x (4M+1), and P be a positive integer smaller than (2M+1) x (2M+1).

Second, block dividing means divides one frame of a television signal into blocks of a predetermined size, and L first blocks existing within an area of ±2M pixels horizontally and ±2N lines vertically. trial vector group storage means for storing a trial vector group of the current frame, motion vector value storage means for storing motion vector values of blocks whose motion vectors have been calculated in the current frame, and motion vectors of blocks whose motion vector values have been calculated surrounding the processing block. a change vector calculation means for calculating a displacement vector using the displacement vector value;
trial vector group selection means for selecting a second P trial vector group distributed within an area of ±M pixels in the horizontal direction and ±N lines in the vertical direction from the trial vector group; Calculation means for calculating the motion vector generated in the processing block of the current frame using the second trial vector group within a range of ±M pixels in the horizontal direction and ±N lines in the vertical direction centered on the processing block. It has been established that However, M and N are positive integers, and L is (4M+1
)X (4M+1), P is a positive integer smaller than (2M+
1) A positive integer smaller than X(2M+1).

Effect The present invention takes advantage of the fact that the motion of a moving object in a video image (for example, a human figure in a TV conference) has a high correlation between successive frames, and uses the first trial vector in which the trial vectors are densely arranged in the center. A second trial vector group is calculated from the group so that the trial vectors are dense in the same direction as the motion vector calculated in the previous frame, and a motion vector search is performed using the second trial vector group. The solution is to perform detection with increased detection accuracy in one step of motion vector search without increasing the number of motion vectors.

In addition, the present invention utilizes the fact that when the processing block size is small relative to the size of the moving object in the moving image, the motion of the moving object in the moving image has a high correlation between adjacent blocks in the same frame, and From the first set of trial vectors located at
A second trial vector group is calculated so that the trial vectors are densely arranged in the same direction as the motion vectors of adjacent motion vector calculated blocks in the same frame, and the motion vector is calculated using the second trial vector group. By calculating, the detection accuracy can be improved by one-step motion vector search with a small capacity of the motion vector value memory section and without increasing the number of trial vectors.

EXAMPLE A first example of the present invention will be described below with reference to FIG. FIG. 1 is a block diagram of a motion vector detection device in a first embodiment of the present invention. In FIG. 1, 1 is a motion vector calculation unit that compares the correlation between pixel values of the current frame and pixel values of the previous frame and calculates a motion vector from the second trial vector group, and 2 is a motion vector that has been calculated. 3 is a trial vector group memory unit that stores the first trial vector; 4 is a trial vector group memory unit that stores the first trial vector; 4 is a trial vector group memory unit that stores the first trial vector;
This is a trial vector group selection unit that selects trial vectors.

The operation of the above configuration will be explained below.

The motion vector calculation unit 1 detects a motion vector using the block matching method described in Conventional Example 1.2 described above. The motion vector calculation unit 1 calculates the current frame pixel value a
The correlation between the pixel values of the previous frame whose positions have been shifted using the trial vector (hHlhv) and the trial vector (hHlhv) is measured for each block using the evaluation function, and the trial vector (hH.

hv) is output as a motion vector (dH, dV).

The internal configuration and operation of the motion vector calculation unit 1 are described in Conventional Example 1.2, and will therefore be omitted.

Next, the motion vector value memory section 2 stores a motion vector (a) at the address indicated by block address C as shown in FIG.
H, dv) and write the motion vector of the previous frame (eH
, eV) memory element 21.2 for one frame
It is composed of 2.

At the start of motion vector calculation for a processing block, the motion vector of the previous frame indicated by block address C (eH).

eV) is read out. Further, when the calculation of the motion vector of the processing block is completed, the motion vector (dH, dv) is written to the address indicated by the block address C. Note that zero vectors are written to all block addresses of the motion vector value memory section 2 before starting motion vector detection processing for the first frame (encoding start frame).

In the following description, the motion vector detection range is assumed to be ±7 pixels in the horizontal direction and ±7 pixels in the vertical direction.

The first trial vector group shown in FIG. 3 is written in the trial vector group memory section 3. The first trial vector group is selected in the horizontal direction ± in order to obtain a second trial vector group within a predetermined motion vector detection range using the motion vectors (eH, eV) of the previous frame in a trial vector group selection unit 4, which will be described later. This is a trial vector group in which trial vectors are arranged in an area of 14 pixels and ±14 pixels in the vertical direction.

FIG. 4 shows the internal configuration of the trial vector group memory section 3.

When a horizontal/vertical address (fHlfV) within the motion vector search area is input from the trial vector group selection unit 4, which will be described later, the address decode circuit 31 converts it into a one-dimensional address. Information g indicating whether the trial vector is valid or invalid is stored in the trial vector value memory circuit 32 for each address information.
Output.

The following table shows part of the relationship between the input address to the trial vector group memory circuit 32 and the trial vector valid/invalid information g for realizing the first trial vector group shown in FIG.

However, when g=1, it is assumed that the trial vector is valid.

Margin below Next, the operation of the trial vector selection section 4 will be explained.

The flow of operation of the trial vector selection section 4 is shown in FIG. 6, and the detailed internal configuration is shown in FIG. When the motion vector value (eH, ev) of the previous frame is input at the start of motion vector calculation of the processing block, the shift vector calculation circuit 41 calculates the (1)
A shift vector (e, eV') is calculated from the formula. [
(a) in Figure 6(a)] e H'== e HX (-
1) --(1) eV=eVx(-1) Next, the detection area calculation circuit 42 calculates the shift vector (e14
', eV'), ±7 pixels in the horizontal direction, ±7 pixels in the vertical direction
A motion vector detection area of 7 pixels is calculated using equation (2). [(O) in Figure 5(a)]]eH-7≦fH≦eH
+ 7 ・・・・・・(2) eH-7≦ fv ≦ e
V+7 The detection area address scanning circuit 43 performs a head-to-head scan of the addresses (fH, fV) in the area calculated by equation (2), and outputs them to the trial vector group memory unit 3. For each address (fH, fV) within the detection area, the trial vector group memory unit 3 outputs information g indicating whether the address (fHlIV) is valid or invalid as a trial vector. [5th
(c) in Figure (a)] The information g and the address information (fH
. Trial vector generation circuit 4
6, a trial vector (hH, hV) is generated by adding the motion vector value e of the previous frame to the candidate address (kH, kV) of the trial vector as shown in equation ('4). a)]h H= k H+ e H--(
3) hV = kv + eV By the above operation, the motion vector value (eH,
FIG. 7 shows the second trial vector group calculated when eV)=(4,4).

Next, a second embodiment of the present invention will be described with reference to FIG. A second embodiment of the present invention will be described below with reference to FIG. FIG. 8 is a block diagram of a motion vector detection device in a second embodiment of the present invention. In Fig. 8, 1 is a motion vector calculation unit that compares the correlation between the pixel values of the current frame and the pixel values of the previous frame and calculates a motion vector from the second trial vector group; A motion vector value memory unit that stores motion vectors of blocks for which motion vectors have been calculated, 2B a shift vector calculation unit that calculates a shift vector using motion vectors of blocks for which motion vectors have been calculated surrounding the processing block, and 3 a first A trial vector group memory section stores the trial vectors of , and 4 is a motion vector selection section that selects a second trial vector group from the first trial vector group using the motion vector of the previous frame.

The operation of the above configuration will be explained below.

However, the internal configuration and operation of the trial vector group memory section 3, the trial vector group selection section 4, and the motion vector calculation section 1 are the same as in the first embodiment, so their explanations will be omitted.

As shown in FIG. 9, the two motion vector value memory sections are comprised of a shift register 2A1 into which the motion vector values of the motion vector calculated blocks adjacent to the processing block are written. At the start of motion vector calculation of the processing block, the 10th
The motion vector values (eIH, eIV) and (e2H, e2V) of two adjacent blocks whose motion vectors have been calculated shown in the figure are read out and output to the shift vector calculation section 2B. When the calculation of the motion vector of the processing block is completed, the motion vector (dH
, dV) are written to the shift register. Note that a zero vector is written into the shift register 2AI before starting the motion vector calculation.

The shift vector calculation unit 2B performs calculation processing of equation (4) to calculate a shift vector (eH, eV).

(e1H+e2H)÷2=eH... (4)(
eIV+e2V)÷2=eV After that, the trial vector selection unit 4 selects the shift vector (eH, eV) by the same operation as described in the first embodiment.
The first vector written in the trial vector group memory section 3 using
Calculate a second trial vector group from the trial vector group of
The motion vector calculation unit 1 uses the second trial vector group to calculate a motion vector (dH).

dV).

Advantages of the Invention As described above, the present invention is useful for moving objects in moving images (for example, TV
Taking advantage of the fact that the movement of objects (images of objects in meetings) has a high correlation between successive frames, we calculate the movement vectors in the same direction as the motion vectors calculated in the previous frame from the first trial vector group in which trial vectors are densely arranged in the center. By calculating the second trial vector group so that the trial vectors are dense and performing motion vector search using the second trial vector group, one-step motion vector search can be performed without increasing the number of trial vectors. Detection can be performed with increased detection accuracy, and the effect is significant.

Furthermore, the present invention utilizes the fact that when the processing block size is small compared to the size of the moving object in the moving image, the movement of the moving object in the moving image has a high correlation between adjacent blocks in the same frame, and A second trial vector group is calculated from the first trial vector group arranged in By calculating the motion vector using the trial vector group of 2,
With a small capacity of the motion vector value memory unit, the detection accuracy can be improved by one-stage motion vector search without increasing the number of trial vectors, and the effect is significant.

[Brief explanation of drawings]

FIG. 1 is a block wiring diagram of a motion vector detection device according to a first embodiment of the present invention, FIG. 2 is a detailed block wiring diagram of a motion vector value memory section in the same embodiment, and FIG. A conceptual diagram showing the first trial vector group,
FIG. 4 is a detailed block diagram of the trial vector group memory section in the same embodiment, and FIG. 6 is a flowchart showing an algorithm for selecting the second trial vector group from the first trial vector group in the same embodiment. FIG. 6 is a block diagram showing the detailed configuration of the trial vector group selection section in the same embodiment, FIG. 7 is a conceptual diagram showing the second trial vector group selected in the same embodiment, and FIG. 9 is a block diagram of a motion vector detection device according to a second embodiment of the present invention, FIG. 9 is a detailed block diagram of a motion vector value memory section in the same embodiment, and FIG. 10 is a processing block diagram of a motion vector value memory section in the second embodiment. Relationship diagram with motion vector detected blocks, 11th
12 are diagrams showing a conventional multi-step motion vector detection method. DESCRIPTION OF SYMBOLS 1... Motion vector calculation unit, 2... Motion vector value memory unit, 3... Trial vector group memory unit, 4... Trial vector group selection unit, 21... Horizontal direction motion vector memory, 22 ...Vertical motion vector memory, 31...
Address decoding circuit, 32... Trial vector value memory circuit, 41... Shift vector calculation circuit, 42...
- Detection area calculation circuit, 43... Detection area address scan circuit, 44... AND gate, 46... Trial vector generation circuit, 2 people... Motion vector value memory section, 2
B9. . Shift vector calculation unit, 2AI...shift register.

Claims (2)

[Claims] (1) Block dividing means for dividing one frame of a television signal into blocks of a predetermined size, and L first blocks existing within an area of ±2M pixels in the horizontal direction and ±2N lines in the vertical direction A trial vector group storage means for storing a trial vector group; a motion vector value storage means for storing a motion vector value of the previous frame; trial vector group selection means for selecting a second trial vector group of P pieces distributed within an area of M pixels and ±N lines in the vertical direction; For each treatment block, use the second set of trial vectors to move ±M horizontally around the treatment block.
A motion vector detection device comprising a pixel and calculation means for calculating within a range of ±N lines in the vertical direction. (However, M and N are positive integers, and L is (4M+1)×(4
M+1), P is a positive integer smaller than (2M+1)×(2
positive integer smaller than M+1) (2) block dividing means for dividing one frame of a television signal into blocks of a predetermined size; and L first blocks existing within an area of ±2M pixels in the horizontal direction and ±2N lines in the vertical direction A trial vector group storage means for storing a trial vector group, a motion vector value storage means for storing motion vector values of blocks for which motion vectors have been calculated in the current frame, and motion vector values for blocks for which motion vector values have been calculated surrounding the processing block. and a change vector calculation means for calculating a displacement vector using the displacement vector value, and ±M pixels in the horizontal direction and ±M pixels in the vertical direction from the first trial vector group using the displacement vector value.
trial vector group selection means for selecting P second trial vector groups distributed within an area of N lines; A motion vector detection device comprising calculation means for calculating within a range of ±M pixels in the horizontal direction and ±N lines in the vertical direction centering on a processing block using a group of trial vectors. (However, M and N are positive integers, and L is (4M+1)×(4
M+1), P is a positive integer smaller than (2M+1)×(2
positive integer smaller than M+1)